Problem Narrative

You’re working on an e-commerce app with 10M+ mobile DAUs. The home feed endpoint returns product cards as JSON, but cellular users in emerging markets are sensitive to payload size: every extra kilobyte increases P95 load time and hurts conversion.

To reduce payload, your backend team proposes a columnar JSON format: instead of repeating the same keys for every item, you send a single fields array and a rows matrix of values. Additionally, to further shrink the payload, you must deduplicate repeated string values (like brand names) into a dict array and replace them with integer indices.

Task

Implement a function that converts a list of JSON-like objects into a compact representation.

Given items (a list of dictionaries) and fields (the ordered list of keys to include), produce a dictionary with:

1. fields: exactly the input fields.
2. dict: an array of unique strings encountered in the selected fields, in order of first appearance.
3. rows: a list of rows, one per item, where each row is a list aligned to fields:
   • If the value is a string, store its index in dict.
   • If the value is an int/bool/None, store it as-is.
   • If a field is missing in an item, store None.

Important Notes

• Only values from the specified fields are considered (ignore other keys).
• Treat True/False as booleans (do not convert to strings).
• Strings are deduplicated globally across all rows.

Examples

Example 1

Input:

• fields = ["id", "brand", "title"]
• items = [{"id": 1, "brand": "Acme", "title": "Socks"}, {"id": 2, "brand": "Acme", "title": "Hat"}]

Output:

• {"fields": ["id","brand","title"], "dict": ["Acme","Socks","Hat"], "rows": [[1,0,1],[2,0,2]]}

Explanation: "Acme" repeats, so it appears once in dict and is referenced by index 0 in both rows.

Example 2

Input:

• fields = ["id", "brand", "price", "in_stock"]
• items = [{"id": 7, "brand": "Zen", "price": 1999}, {"id": 8, "price": 2499, "in_stock": False}]

Output:

• {"fields": ["id","brand","price","in_stock"], "dict": ["Zen"], "rows": [[7,0,1999,null],[8,null,2499,false]]}

Explanation: Second item is missing brand, so None is placed in that position.

Constraints

• 1 <= len(items) <= 2 * 10^5
• 1 <= len(fields) <= 50
• Keys in fields are unique
• Values for selected fields are one of: str, int, bool, None
• Total number of selected values across all items is up to 10^6

Goal

Return the compact representation efficiently (linear in the number of selected values).